The dicatphone was revolutionary.
By the 1950s, a new sound was becoming familiar in law offices, the click of a dictation machine.
Instead of sitting down to write a letter, a lawyer could speak it into a recorder. The tape would then pass to a secretary, who would type it up, return a draft and eventually prepare the finished document. To modern eyes, it looks almost comically analogue. At the time, it was productivity technology.
Its promise was straightforward. Lawyers were expensive; their time was scarce. If technology could remove some of the administrative work surrounding legal advice, they could spend more of their day doing the work that supposedly required a lawyer.
Seventy years later, legal AI is being sold with remarkably similar logic.
The machines have changed rather more than the sales pitch. Dictaphones gave way to word processors. Filing cabinets were supplemented by databases. Letters became emails. Search engines began retrieving cases in seconds. Machine learning started sorting millions of documents. Then chatbots learned to draft them.
Now the industry is moving towards AI agents that promise not merely to help lawyers complete individual tasks, but to carry parts of a legal workflow from beginning to end.
The first wave of automation addressed administrative functions
Long before generative AI, lawyers sought to routinize and automate elements of the law that seemed impediments to productivity. In the 1950s, law firms invested in dictation machines, allowing lawyers to dictate correspondence to be transcribed by secretaries later. It is not the same thing, but the economic rationale is familiar to modern legal-tech entrepreneurs: to extract as much of the drudgery as possible from the law so that more work can get done in the same amount of time.
The next innovation, the personal computer, arrived in the 1980s and was complemented by the rise of legal-specific software for managing accounting, timekeeping, and case management. Local-area networks allowed documents and printers to be shared, and word-processing software gradually replaced typewriters.
The automation of legal work was even more profound, though perhaps less visible, in the 1990s. As computers grew more powerful and more widespread, the ability to store, retrieve and manipulate information became more sophisticated.
Search was the first superpower
Legal research was a rather obvious application. Before computers, the task of finding the law was a laborious and uncertain process. It typically involved wading through books and pamphlets to find relevant statutory language or court decisions. With the rise of electronic databases, searching became easier and more precise. In 1973, Mead Data Central began selling the LEXIS product suite, giving lawyers access to research databases containing Ohio and New York cases, statutes, federal caselaw, and the US Code. In 1974, lawyers could perform searches remotely over a telecommunication network.
This has since been superseded by more modern research databases. Nevertheless, it is worth recognizing just how momentous this innovation was to the practice of law. For the first time, the information needed to navigate the law could be retrieved directly.
At around the same time, researchers began asking if computers could go beyond automating research and actually perform some legal reasoning tasks as well. In the 1970s, Thorne McCarty created TAXMAN, one of the first legal expert systems, designed to assist with tax law. Between 1980 and 1987, legal AI research blossomed, with various programs using rule-based reasoning dominating the field. The International Conference on Artificial Intelligence and Law was first held in Boston in 1987.
Legal AI was not limited to narrow applications such as research or tax law. Still, these early systems suffered from limitations in their ability to represent the law in a way that could be easily understood by a computer.
Meanwhile, artificial neural networks have advanced by leaps and bounds, enabling modern AI to perform tasks that would have been impossible a few decades earlier.
Then came machines that could read
The breakthrough that enabled modern AI was not in teaching it to reason like a lawyer but rather to recognize patterns in data. Electronic discovery was one of the first applications, and it continues to play a critical role in legal AI. The process of reviewing documents during litigation can be grueling, especially for large volumes of material. It is not uncommon for cases to generate more than three million files that need to be reviewed. Paying people to comb through this information is time-consuming and expensive. Various machine-learning algorithms have been developed to identify relevant documents and reduce the burden on humans. In 2012, the case of Da Silva Moore v Publicis Groupe became the first to formally use predictive coding to review documents — and the first US judicial opinion to approve of it.
This is unquestionably a form of artificial intelligence. Even so, it plays a relatively narrow role, assisting humans in a specific task. The user is not conversing with an AI but rather providing it with information that the algorithm then uses to perform useful computations.
The chatbot changes the interface
Long before ChatGPT, legal chatbots have been promising to change the legal world. DoNotPay, a service initially designed to help users dispute parking tickets, was launched in 2015 and branded as a robot lawyer. Its founder later expanded its capabilities to address a range of other issues. Not surprisingly, the FTC launched an action against the company in 2025, citing false advertising by claiming that its service could act as a substitute for a human lawyer.
Many legal chatbots were essentially portals to more traditional legal services. A user could select an option and respond to a few prompts, and the chatbot would guide them to the relevant information or documents. That is undoubtedly useful, but how intelligent was it really?
Modern large language models have dramatically exceeded the capabilities of their predecessors.
Just….write
OpenAI’s release of ChatGPT to the public in late November 2022 marked the beginning of a new era. Suddenly, everyday users found themselves able to issue instructions to a computer and receive coherent responses.
For legal tech, this was an important development because it meant users no longer had to learn how to interact with the software in a particular way. They could simply speak to it like another person.
New specialized legal products emerged throughout 2023. In March, Casetext announced the launch of CoCounsel, a product suite utilizing GPT-4 to assist lawyers with document review, memo creation, deposition preparation, and contract analysis. Later that year, LexisNexis announced a commercial preview of Lexis+ AI. Thomson Reuters then acquired Casetext for $650 million and integrated generative AI into its broader product suite.
The importance of this development is difficult to overstate. For the first time, lawyers gained access to truly powerful AI tools capable of performing a wide range of tasks.
A legal memo might be created using natural language prompts. Summaries can be generated automatically. Authorities can be cited. Similar laws and contracts can be compared. An entire research memo might be produced based on a few questions. The possibilities are virtually endless.
From answers to actions
Modern developments in legal technology seek to capitalize on this capability. The preferred term is agentic AI — and the distinction is an important one.
Whereas chatbots typically respond to prompts, an agent is capable of doing more: it can formulate goals, identify the necessary steps, interact with various elements, and achieve outcomes. The idea is to enable computers to take on more substantial roles in performing legal tasks. Thomson Reuters began referring to its own CoCounsel platform in agent-friendly language last year, and more recently, Google launched Gemini Enterprise for Legal in August 2026. The API connects AI agents with legal and enterprise software, enabling them to perform more end-to-end functions throughout a legal workflow.
It should be noted that agentic AI does not necessarily mean self-driving robot lawyers that can operate independently. Issues of reliability, confidentiality, professional responsibility, and verifiability limit the extent to which such an outcome is realistic in the near future. At best, AI agents are likely to supplement lawyers by performing discrete functions.
With that said, the trajectory of innovation appears indisputable.
For more than 70 years, legal technology has steadily advanced toward the core functions performed by lawyers. Dictation machines captured the lawyer’s voice. Word processors helped to produce it. Databases assisted in finding it. Machine learning helped to sort through it. AI helped to create it. Agentic AI now aims to organize it all.
That is why the current moment appears so momentous. Most previous forms of legal technology were designed to make individual tasks easier. Agentic AI seeks to make entire processes more efficient.
There is reason to be skeptical of the more apocalyptic predictions regarding the demise of lawyers. After all, the profession has seen one supposedly disruptive innovation after another over the past century.
Still, history provides one lesson that remains relevant: lawyers used to argue about whether email would be appropriate for legal practice. They no longer do.
